ICONS 2018 : The Thirteenth International Conference on Systems System Performance Condition Curve Estimation Based on Data Analysis of the Taipei Metro System Tzu-Chia Kao and Snow H. Tseng* Department of Electrical Engineering, National Taiwan University, Taipei 10617, Taiwan e-mails: [email protected], [email protected] Abstract—The Taipei Metro is still very young compared to the condition via in-service vehicles [3-5]. The general goal of other metropolitan metro systems in the world. The data the research and technical modifications is to improve the provided by Taipei Rapid Transit Corporation (TRTC) is performance and reliability of a mass rapid transit system. analyzed, including: air conditioner, communication system, switcher, platform door, 22kV switchboard, Automated fare We also looked into some older subway systems on the collection (AFC) door, Wenhu Line traffic control computer, Internet. As reported in [6], train R36 serviced from 1964 to Transmission system, elevator and escalator. The research 2003, a total of 39 years. R160s were used to replace 45-year- objective is to assess the performance based on the one-year- old trains. In another news report about the old trains [7], the long data from TRTC, to determine the equipment stage of its oldest trains for New York City Subway were planned to serve lifetime, and estimate the asset’s remaining life and for 58 years, and now this type of trains are actually found too maintenance condition. Reported analysis of the data shows old with very high failure rate and not appealing to meet minute indication of the system current status; further analysis passengers. From the limited reference that we can access, an is in progress. estimate of the subway train lifetime is estimated to be around Keywords-degradation; maintenance; metro; MRT; performance 40 to 50 years. For example, some lines of Singapore Mass analysis. Rapid Transit (SMRT) have been operating since 1987, 30 years from today. On the other hand, TRTC operated from I. INTRODUCTION 1996, which is only 21 years ago. There is a difference of 9 years. The assets’ actual wear-out period may lie somewhere We analyze the maintenance record of the Taipei Mass Rapid Transit (MRT) system, which is a well maintained between 20 years (the oldest TRTC asset), and 40 years (New York City Subway). All these metropolitan metro systems are system renown worldwide for its tidiness, stability and efficiency. If the maintenance data can be analyzed to yield different in various aspects, thus, the characteristics of these MRT systems are expected to differ; the predictability and information about how to perfect the performance of the Taipei MRT system, the analysis can provide useful accuracy of the estimation and extrapolation based on TRTC to assess other MRT systems using the TRTC system data is information to improving Taipei MRT, or, more generally, to assess the performance of other systems. understandably incomplete; with unknown number of variables involved, the accuracy of assessment may be very Since the MRT system of Taipei Rapid Transit limited. Corporation (TRTC) is relatively young (began operation on Assessment and quantification of the system current March 28, 1996, a total of 22 years to date [1]), most of the status is essential to enhance performance. The degradation equipment has not yet been replaced; the records of repair and curve is commonly employed for estimation of the system maintenance are detailed and are stored in digital form. Thus, current status. Analysis of the reliability is commonly based our research goal is to establish a degradation curve based on on failure rate and maintenance records [8]. Based on the the maintenance and performance record, and use to assess the status of the system, possible improvement of the Taipei MRT system, and equipment of other metro systems. maintenance and performance can be assessed. By means of The performance and degradation of metropolitan metro the degradation analysis the research objective is to shed light systems have been the focus of general public. Various on enhancing the metro system performance. studies have been reported, including technical issues of the To study the maintenance and performance MRT. Rail track condition monitoring is an important characteristics, various approaches have been reported [9-18], technical concern of the MRT system [2]. However, it is including the popular bathtub curve analysis [19]-[24]. The infeasible to constantly inspect track conditions; an Bathtub curve is commonly employed for system inspection once a month or less is more or less the usual performance analysis [25]. Analysis based upon the bathtub maintenance. Severe track condition degradation that isn’t curve has been extensively applied to various problems; detected early is a potential threat to the railway system. various modifications to improve applicability have been Hence, more attention has been devoted to monitoring track reported [11][26]-[29]. It is suggested that bathtub curve Copyright (c) IARIA, 2018. ISBN: 978-1-61208-626-2 25 ICONS 2018 : The Thirteenth International Conference on Systems could be interfered by human factor [30]; for example, if the asset retired in its early stage, the curve may not rise up during the wear-out period and may even descend. If properly maintained, the curve may not rise in the wear-out period, similar to the situation in airline industry. However, few MRT Third, average each equipment’s failure times to get the systems in reality exhibit degradation behavior similar to the average failure rate. Use a statistical software to calculate the bathtub curve model [31]. regression curve, and extrapolate for comparison with other The rest of this paper is organized as follows. Section II metro system performance data. describes the goal of this research project. Section III Continue these steps to process another set of data describes the research method. Section IV summarizes the accordingly; then compare the regression behavior. Analyze data analysis; finally, a summary is presented in Section V, the difference between the two datasets. followed by an acknowledgement. IV. DATA ANALYSIS Initial attempt of analysis is performed. We use the failure II. RESEARCH GOAL statistics for monthly failure rate provided by TRTC. The We proposed to thoroughly analyze the Taipei MRT data. AFC gates statistics is shown in Figure 1; the averaged We have data from Taipei MRT consisting of 11 systems: malfunction rate is calculated and shown in Figure 2. As electric multiple unit (EMU), EMU air conditioner, EMU shown in Figure 1, the equipment consists of mixed brand, communication, switcher, platform door, 22kV switchboard, model, age; thus, the total number of malfunctions is not Automated fare collection (AFC) door, Wenhu Line traffic representative of individual equipment. control computer, Transmission system, elevator, and escalator. By analyzing the dataset with various state-of-the- art approaches, such as deep learning, the research objective is to analyze the equipment present status and identify possible tendencies or features that may be indicative of the system performance. III. METHOD Ideally, analysis of the maintenance data would yield a simple bathtub-shaped degradation curve for each equipment. However, the bathtub-shaped degradation curve is a theoretical model; degradation curve of most cases do not follow the same degradation curve, not to mention each equipment may exhibit different characteristics. Furthermore, each equipment in the Taipei metro system consists of various brands and various models that may possess different intrinsic characteristics. Furthermore, since each equipment is maintained by human, the degradation curve is unlikely to be a simple universal bathtub-shaped curve. Thus, it is infeasible to come up with a universal bathtub-shaped degradation curve for each Taipei MRT equipment. Based on the experience we Figure 1. Failure rate of AFC gates vs. time. The failure rate decreases monotonically with time. had from the initial attempt to analyze the data, we propose the following steps: First, based upon the original maintenance records, calculate the duration between: 1) when the equipment was first engaged in operation, and, 2) the failure date. This time Based upon the trend of the data, the failure rate has been interval represents the duration of malfunction-free operation, declining yearly until it reaches low and stable end tail. which is also the time for a malfunction to take place. By Possible factors are speculated, this mostly likely is due to the analyzing the malfunction-free duration instead of the number improvement of MRT maintenance. On the other hand, since of malfunctions each month may yield more realistic the age of each equipment and the number of samples for each relationship. equipment are not consistent, the total failure rate is not a fair Second, calculate the failure times per month to acquire representation of a specific individual equipment. Thus, we the failure rate for each individual equipment. (For the target to analyze the age’s effect of the individual equipment equipment related to EMU, the failure rate is acquired by the as far as possible. failure time divided by the train mileage.) Copyright (c) IARIA, 2018. ISBN: 978-1-61208-626-2 26 ICONS 2018 : The Thirteenth International Conference on Systems supported by the Taiwan National Science Council Grant 106-2112-M-002-008 and National Taiwan University grant NTU-ERP-103R89086. REFERENCES [1] "Taipei Metro," Wikipedia. [2] X. K. Wei, F. Liu, and L. M. Jia, "Urban rail track condition monitoring based on in-service vehicle acceleration measurements," Measurement, vol. 80, pp. 217-228, Feb 2016. [3] M. Molodova, M. Oregui, A. Nunez, Z. L. Li, and R. Dollevoet, "Health condition monitoring of insulated joints based on axle box acceleration measurements," Engineering Structures, vol. 123, pp. 225- 235, Sep 2016. [4] G. Lederman, S. H. Chen, J. Garrett, J. Kovacevic, H. Y. Noh, and J. Bielak, "Track-monitoring from the dynamic response of an operational train," Mechanical Systems and Signal Processing, vol.